Accurate and scalable surface representation and reconstruction from images

Area of Science:

• Computer Vision
• Computer Graphics
• Geometric Modeling

Background:

• Current 3D surface reconstruction methods often struggle with objects of arbitrary size or complexity.
• Existing techniques may be limited to either open or closed surfaces, not both.
• Efficient and detailed surface reconstruction remains a key challenge in computer vision.

Purpose of the Study:

• To introduce a new surface representation, "patchwork," for enhanced 3D surface reconstruction from multiple images.
• To enable reconstruction of objects with arbitrarily large dimensions while maintaining fine detail.
• To develop a method that handles both open and closed surfaces uniformly.

Main Methods:

• The patchwork representation combines multiple patches, built incrementally.
• Formal analysis demonstrates spatial complexity independent of object dimensions and linear time complexity with respect to object area.
• The method is shown to be orthogonal to surface optimization techniques, integrating with existing approaches like minimal cuts.

Main Results:

• Patchwork achieves storage efficiency and reasonable reconstruction time.
• The representation successfully reconstructs both open and closed surfaces.
• Demonstrated applications include enhanced carving techniques and surface propagation for detailed geometry, handling complex topologies.

Conclusions:

• The patchwork method offers a versatile and flexible approach to 3D surface reconstruction.
• It overcomes limitations of existing methods in handling object scale and surface type (open/closed).
• Patchwork representation enables robust reconstruction of complex shapes and topologies from image sequences.